Experimental Evaluation of MIMO Observers and Model Based Controllers

Abstract

State estimators for MIMO systems and model based controllers are important for multivariable control. Many publications have presented results of multivariable control, however most of the results are obtained from simulations. In this thesis the performance of Kalman filters and model based controllers was evaluated experimentally using the pilot sized quadruple tank process located at USN department of Porsgrunn. Careful development and calibration of different process models was considered in order to obtain reasonable results from the experimental evaluation. The Kalman filters and model based controllers were subject to deliberate modeling errors to evaluate robustness to model mismatch. Models were obtained from first principles modeling and subspace system identification using the DS-R method for model realization. Experiments showed that the Kalman filters produce estimation error in the internal states while accurately estimating the outputs. The model based controllers showed to be robust to modeling errors due to feedback.